About agentic-engage
agentic-engage is an open-source engagement queue for Node.js developers and agents, with approval controls, local receipts, and a generic HTTP adapter. Social-platform labels are provided, but native platform delivery is not implemented; actual delivery depends on a caller-supplied endpoint.
Key features
- 01Draft engagement text from fixed templates
- 02Approve or deny actions before execution
- 03Deduplicate targets and keep an in-memory audit log
- 04Dispatch through a caller-configured HTTP endpoint
- 05Apply optional caps, quiet hours, and blocklists
Use cases
- 01Prepare engagement actions for human approval
- 02Track planned and completed actions in a local queue
- 03Send approved text to a supplied webhook endpoint
- 04Limit dispatches during quiet hours
MCP support
The MCP server exposes tools to plan engagement, inspect the queue, approve or deny actions, execute them, and retrieve a cost report. MCP execution uses local receipts and cannot configure HTTP adapters.
Configuration
Local server config is the same for every client. Copy it into your client’s MCP config file.
Tools
6Saved tool list · updated 2w ago
| Parameter | Type | Required | Description |
|---|---|---|---|
| targets | array | Yes | |
| personaVoice | string | Yes |
Details
From the project READMEagentic-engage
agentic-engage is an open-source engagement queue for Node.js developers and agents, with approval steps, local receipts, and a generic HTTP adapter.
Install
npm install agentic-engage
Requires Node.js 18 or later. Includes TypeScript declarations and two runtime dependencies.
Preview your first action
Save this as preview.mjs, then run node preview.mjs. No credentials are needed.
import {
createEngagementQueue, planEngagement, approveEngagement,
executeEngagement, getCostReport,
} from "agentic-engage";
const queue = createEngagementQueue();
const [draft] = planEngagement(
[{ id: "post-1", platform: "webhook", context: "your launch" }],
"Helpful brand", queue,
);
approveEngagement(draft.id, queue);
const preview = await executeEngagement(draft.id, { queue });
console.log(preview.dryRun, preview.action.status); // true approved
console.log(getCostReport()); // entries: [], dispatches: 0, units: 0
Success prints true approved and zero dispatch costs. Preview makes no delivery request.
Features and coverage
- Drafts: Generate fixed template text with an AI-assistance disclosure.
- Review: Approve or deny actions; execution requires approved status.
- Queue: Deduplicate target IDs and keep an in-memory audit log.
- HTTP adapter: POST target and text to an endpoint you supply.
- Guards: Optional dispatch caps, UTC quiet hours, and text blocklists.
| Target label | Native integration status |
|---|---|
| scaffold; generic HTTP or local receipt only | |
| X (Twitter) | scaffold; generic HTTP or local receipt only |
| scaffold; generic HTTP or local receipt only | |
| YouTube | scaffold; generic HTTP or local receipt only |
| Bluesky | scaffold; generic HTTP or local receipt only |
| Webhook | Generic HTTP adapter implemented; endpoint supplied by caller |
Six target labels share two engagement adapter classes. No native social API is credential-verified.
Use new HttpEngagementAdapter("webhook", { endpoint }) in the library's adapters map.
Your endpoint handles authentication and platform delivery. No vendor environment variables or .env loader exist.
Runtime surfaces
CLI commands: plan, queue, approve, deny, send, costs, and mcp; also --help and --version.
Run npx agentic-engage mcp or npx --package agentic-engage agentic-engage-mcp for Model Context Protocol (MCP).
Tools: plan_engagement, get_engagement_queue, approve_engagement, deny_engagement, execute_engagement, and get_cost_report.
Worker source exposes GET /health, POST /summarize, and POST /plan; you deploy it yourself.
Limits and status
Queues and costs reset when the process exits. Separate CLI invocations do not share actions.
CLI and MCP live execution use local receipts; they cannot configure HTTP adapters.
Library execution also defaults to local receipts without an adapter, even with live: true.
A sent status therefore does not prove social publication. HTTP receipts use locally generated IDs.
HTTP requests default to a 10-second timeout with no retries. Costs count characters, not currency.
Drafts use fixed text templates; no language model generates them. Account ownership is not verified.
Version 0.1.0 has an API that may change. No managed social delivery service is included.
Links and license
Comments
New to MCP?
From what the protocol is to installing your first server, these answers cover most first questions.
An open protocol that defines how AI applications connect to external tools and data in a uniform way. Clients talk to the model; MCP servers provide tools, resources and prompt templates. They exchange standard messages, so one server can be reused by many clients.